Table 2 Table of training parameters
Hyperparameters | Value | Description |
|---|---|---|
max_epochs | 900 | Total number of epochs for the training phase |
num_gpus | 1 | Number of gpus used |
up_to_num_nodes | 353996 | Total number of nodes contained in the subgraph we sample (in this downsized situation, we decide to sample a subgraph which is almost as big as the full graph) |
num_start_nodes | 1 | Number of randomly sampled source nodes for BFS explorations (subgraph sampling): if number == n, it means that n independent BFS will be performed and the resulting subgraphs aggregated in a sinle edge list |
max_steps | 90 | Total number of training iterations per epoch |
accum_steps | 5 | Number of gradient accumulation steps |